JP3320320B2 - A buffer management method using a buffer lock technique in a multi-user environment storage system - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

TECHNICAL FIELD The present invention relates to a multi-user (mult
In i-user) environment storage system, comprising relates buffer management method using the locking techniques buffer, treated in particular simultaneous access of many users for the same disk page (concurrent access) efficiently, This is to maximize the effect of buffer management used to improve the performance of the storage system.

[0002]

2. Description of the Related Art A buffer is a memory space used to hold frequently used data among data on a disk, which is a physical storage space, in a memory.

[0003] Among the modules constituting the storage system, a buffer management module is a module for managing the space of a buffer allocated in a memory, and a physical storage space is frequently used in disk data. In a buffer to enhance the overall performance.

[0004] The main function of such a buffer manager, immediately A page of disks that are commonly used
Access read into the buffer of the memory can be a page held in the buffer memory system recovery Ma
In conjunction with the Recovery Manager , this means properly flushing the relevant page on the disk and minimizing disk access for data processing.

[0005] Buffers are one of the most frequently accessed resources in storage systems,
The buffer manager 's efficient buffer management scheme minimizes disk I / O for data processing, thereby improving the overall performance of the storage system.

In the prior art, when accessing a page to process a user's request, if the image of the page does not exist in the buffer, the page should be read from the disk to the buffer for data processing. It is.

Such a request can be made simultaneously by a number of users using the storage system, wherein the operation of reading a page from disk to a buffer is repeated by the user's request. Can be.

However, when an operation for reading the same page from a disk to a buffer is performed simultaneously by a large number of users, a large number of buffers are used repeatedly for the page of the same disk, and the following problem occurs. I do.

First. 1. Waste of buffer space occurs. Unnecessary disk I / O occurs.

Third. Buffer replaceme
nt), and an overhead of 4th. An inconsistency occurs between the contents of the buffer and the contents of the disk page.

In order to prevent the system performance from deteriorating due to such a problem, it is necessary to perform processing so that disk input / output for the same page is not duplicated.

Typical methods conventionally used to solve this problem are classified into the following two types.

First. An exclusive latch (exclusiv) is attached to a buffer hash anchor (Buffer Hash Anchor) that manages the corresponding buffer to access a desired page.
e latch) is applied (acquire), and a buffer having an image of a desired page is searched (scanned) from a buffer frame chain (Buffer Frame Chain).

If the corresponding buffer exists in the buffer frame chain, the determined buffer is exclusively latched, and the next buffer which has released the exclusive latch of the buffer hash anchor is used.

[0015] If the desired page is not present in the buffer frame, the buffer hash anchor hangs the desired page from the disk without releasing the exclusive latch on the buffer hash anchor so that requests for the same page by other transactions are not processed. After the buffer is connected to the buffer frame chain in the locked state, the exclusive latch of the buffer hash anchor is released.

After the page is allocated from the disk and read into the buffer, the exclusive latch for the buffer is released and the buffer is used.

The first method described above, ie, allocating a free buffer and continuing to latch the buffer hash anchor while coupling the buffer to the buffer frame chain, is simply the same. This has the advantage that duplicate disk I / O for pages can be implemented.

However, in this method, while the buffer hash anchor is latched, another transaction that attempts to access a page other than the page cannot access the corresponding buffer frame chain.

Therefore, when this method is used, when many transactions request an empty buffer at the same time, the exclusive latching time for the buffer hash anchor is increased, and the performance of the system is sharply increased. There is a problem of being reduced.

Second. In order to access a desired page, it has a lock on the page in an exclusive mode.

Once the lock on the page has been executed,
Access the buffer hash anchor.

If a buffer including the image of the corresponding page exists in the buffer frame chain, the lock on the corresponding page is released using the buffer.

If there is no buffer corresponding to the page in the buffer frame chain, an empty buffer is allocated, an exclusive latch is applied to the buffer hash anchor, and the allocated empty buffer is stored in the buffer frame chain. After coupling, release the exclusive latch of the buffer hash anchor.

After the page is read from the disk into the buffer, the use of the buffer is terminated and the lock on the page is released.

The above-described second method, that is, a method of holding a page lock until the use of a desired page is finished, is not only easy to implement, but also while reading the page into a buffer. , The execution of another transaction requesting another page as in the first method does not occur.

However, each time a page is searched, the page must be locked. Therefore, an overhead of maintaining a table of locks for all pages accessed by a transaction occurs, and exclusive use is performed every time a buffer is used. Having one lock for each mode causes a decrease in system performance.

[0027]

As described above, when the first method is used, when many transactions request an empty buffer at the same time, an exclusive latch is applied to the buffer hash anchor. time increases, problems performance of the system is reduced rapidly occurs, when using the second method, each time looking for pages, since it is necessary locks to a page, transaction access There is a problem that the overhead of holding a lock table for every page to be generated occurs, and having one exclusive mode lock each time a buffer is used causes a decrease in system performance. .

The present invention has been made to solve these problems. When a buffer manager accesses a page of data to process a user's request, the buffer manager accesses the corresponding page. If the image is not in the memory buffer, the page must be read from disk to the buffer, and in such a case, unnecessary disk insertion is performed without interfering with the execution of other transactions that access other pages. Multiple use that can improve the performance of storage systems from multiple user environments by providing a buffer management method that can efficiently process transactions that require access to the same page at the same time without output Management method using buffer lock technique in storage system of user environment For the purpose of obtaining.

[0029]

A technical feature of the present invention to achieve the above object is that when a transaction accesses an arbitrary page, the transaction exclusively latches the buffer hash anchor by the transaction. A first step of determining whether there is a desired buffer for the corresponding page from the corresponding buffer frame chain; and a buffer lock chain if no buffer for the desired page is found from the corresponding buffer frame chain by the first step. A second step of determining whether a buffer lock entry for the desired page exists in the buffer lock chain, and a case where a buffer for the desired page is found from the buffer frame chain by the first step. ,buffer After the exclusive latch of the buffer hash anchor is released and the exclusive latch of the buffer hash anchor is released, the buffer lock entry for the desired page is determined by the third step of using the buffer page and the second step. If present in the chain,
A fourth step of registering the transaction information in the channel of the corresponding buffer lock entry, releasing the exclusive latch of the buffer hash anchor, waiting, and then exclusively latching the buffer hash anchor; and If the buffer lock entry for the desired page does not exist in the buffer lock chain according to the determination of the process, a fifth lock for the page is executed.
To perform the process.

That is, in a buffer lock method for efficiently managing the simultaneous access of multiple users to the same disk page and improving the performance of the storage system, a buffer lock entry (buffer loc) is provided.
Allocation of free buffers to read pages from disk into buffers using the buffer lock chain, which is a list form of k entry)
Does not use an exclusive latch on the buffer hash anchor while requesting the other, by performing a lock on the page read from disk only if the buffer for the desired page from the relevant buffer frame chain is not found. It prevents the transaction that requires access to the page from being executed and prevents other transactions that request allocation of a free buffer at the same time from using the exclusive latch for the buffer hash anchor for a long time, thereby preventing the entire system from being used. Prevent performance degradation, lock only pages read from disk, minimize locking overhead, efficiently manage buffers without unnecessary disk I / O, and improve storage system performance Characterized by

[0031]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Preferred embodiments of the present invention will be described below in detail with reference to the accompanying drawings.

FIG. 1 shows a hardware environment to which the present invention is applied.

In FIG. 1, a processor [Pi (i = 1
To n)], a main storage memory 12, a shared memory area (shared memory area 13) in the memory, an input / output processor (IOP) 14, and a disk 15 are disks of an auxiliary storage device.

Thus, the present invention can operate from a general hardware environment including a single processor or multiple processors and a shared memory area.

FIG. 2 shows a module structure of a storage system in which the present invention is operated.

The storage system to which the present invention is applied is basically an interface management module 22 for connecting a user 21 to the system, and a data storage / manipulate that creates and processes a database 25 on a disk.
The access management module 23 includes a transaction management module 24 that is in charge of transaction concurrency control and recovery.

[0037] A buffer manager using the buffer locking mechanism of the present invention .
Manager is a data storage / access management module
23 .

FIG. 3 shows an object used for buffer management in a storage system to which the present invention is applied. In this drawing, only parts necessary for explaining the present invention are shown.

The buffer 31 that has read the disk page at the request of the user is connected to the buffer frame chain 33 of the corresponding buffer hash 32 while having a leading disk page image in the buffer.

The buffer connected to the buffer frame chain may be such that the page contained in the buffer is no longer valid or the buffer is replaced by a buffer replacement algorithm.
Retained until you have an image of another page.

Therefore, if a buffer for a desired page is not found in the corresponding buffer frame chain, the page is not in the buffer and should be read from the disk, so that the buffer pool (buffer pool, 34) Free buffer from (em
pty buffer) is newly allocated.

The buffer lock method of the present invention manages buffers while maintaining a buffer lock chain, thereby preventing the current situation of requesting the same disk page redundantly for buffer allocation, and preventing other pages from being allocated. This is a method of managing buffers efficiently without interrupting the execution of other transactions that access .

FIG. 4 shows the structure of a buffer lock chain used in the buffer lock method of the present invention.

The structure of the buffer lock chain is such that the buffer lock entry 41 has a list form (list structr).
ue), the structure of each buffer lock entry that makes up the buffer lock chain
Is It is.

In the buffer lock entry structure,
PageID is a page identifier of a page to be read from the disk, next is a pointer to the next buffer lock entry, and channel is a list indicating a transaction 42 waiting for the PageID page to be read from the disk. is there.

The buffer lock entry is used to prevent a current transaction from allocating a buffer to the same page and performing a disk read operation when other users request access to the same disk page at the same time. The transaction that performs the read operation locks the relevant page (hol)
Used per d).

That is, a transaction for performing a disk read operation constructs a buffer lock entry for a corresponding page, and then connects the entry to a buffer lock chain.

The transaction that has read all pages from the disk to the buffer releases the lock on the page by deleting the corresponding buffer lock entry from the buffer lock chain.

FIGS. 5 and 6 are flowcharts showing a buffer management method using a buffer locking method in a page request processing process according to the present invention.
Note that A in FIG. 5 is connected to A in FIG.
B in FIG. 6 and B in FIG. 6 are connected, and C in FIG. 5 and C in FIG. 6 are connected.

When a transaction accesses an arbitrary page,
When the scan, the transaction multiplied by the exclusive latched into buffer hash anchor (acquire) (S1), find the buffer for the corresponding page from the relevant buffer frame chain (S2).

When the buffer is found, it is determined whether there is a desired buffer (S3). If there is a buffer for the page in the buffer frame chain, the buffer is exclusively latched (S4) and the buffer hash anchor is set. Unlocked exclusive (re
After the lease (S5), the page of the buffer is used (S6).

However, if there is no buffer for the page in the buffer frame chain, the buffer lock chain is inspected while holding the exclusive latch of the buffer hash anchor (S7).

The buffer lock chain is checked to determine whether there is a desired page (S8). If there is a desired page, if a buffer lock entry for the desired page exists in the buffer lock chain, another transaction is performed. Since the relevant page is being read from the disk to the buffer, this transaction is a transaction waiting for the relevant page to be read into the buffer by registering the transaction information in the channel of the buffer lock entry for the relevant page. Is displayed (S
9) The exclusive latch of the buffer hash anchor is released (S10), and after waiting (Block) (S11), the wake-up is performed and the process is re-executed from S1.

The buffer lock chain is checked to determine whether there is a desired page (S8). If the information on the desired page does not exist in the buffer lock chain, this means that another transaction requesting the corresponding page is performed. Means that there is currently no buffer lock entry for the page so that this transaction can allocate buffers and wait for another transaction's request to complete the operation of reading the page from disk. Is newly constructed (S1
2) Register (or connect) to the buffer lock chain
(S13).

Then, the exclusive latch of the buffer hash anchor is released (S14). After a free buffer is allocated (S15), a desired page is read from the disk into the buffer (S16).

Thereafter, the buffer hash anchor is exclusively latched (S17), and the buffer is connected to the buffer frame chain (S18).

A buffer lock entry for the read page is found from the buffer lock chain (S
19), the channel is separated by the buffer lock entry (S20), and the lock on the page is released by deleting the buffer lock entry from the buffer lock chain (S21).

The exclusive latch of the buffer hash anchor is released (S22), and all transactions connected to the separated channel are woken up (wa).
keup), put into an execution state (S23), and execute S5 and S6.

Waking up from the waiting state (wakeu
p), the transaction in the execution state is re-executed from S1.

The present invention as described above has the following effects.

The present invention allocates a new buffer and reads the page from disk to the buffer while the page required for use from the prior art does not exist in the buffer. To solve the problem that the performance of the system was degraded when a large number of transactions requested the allocation of an empty buffer at the same time when a transaction requesting another page was not executed when latching. Does not prevent the execution of a transaction requesting another page of the same hash value.

Further, it is possible to prevent the performance of the entire system from deteriorating due to a transaction requesting allocation of a free buffer.

According to the present invention, when a lock is held on a page every time a page is requested in the prior art, an overhead of holding a lock table for all pages accessed by a transaction occurs, and a buffer is allocated. Each time it is used, the lock of the mode is exclusively held, and by solving the problem that the system performance has been degraded, the lock is executed on the page only when there is no buffer for the page. , Minimizing overhead for locking.

In addition, by preventing unnecessary disk input / output, the effect of buffer management used for improving the performance of the storage system is maximized.

[0065]

According to the first aspect of the present invention, the performance of the entire system can be prevented from deteriorating by not hindering execution of a transaction requesting access to another page.

According to the second aspect of the present invention, locking is performed on a page only when there is no buffer for the corresponding page, so that the overhead for locking can be minimized.

According to the third aspect of the present invention, it is possible to prevent a decrease in the performance of the entire system due to a transaction requesting allocation of a free buffer.

According to the fourth aspect of the present invention, by preventing unnecessary disk input / output, the effect of managing the buffer used for improving the performance of the storage system can be maximized.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of hardware to which the present invention is applied.

FIG. 2 is a module configuration diagram of a storage system to which the present invention is applied.

FIG. 3 is a diagram illustrating a buffer management structure in a storage system to which the present invention is applied;

FIG. 4 is a structural diagram of a buffer lock chain used in the buffer lock method of the present invention.

FIG. 5 is a flow chart showing a buffer management method using a buffer lock method in a process of processing a page request according to the present invention;

FIG. 6 is a flow chart showing a buffer management method using a buffer lock method in a process of processing a page request according to the present invention.

[Explanation of symbols]

1 Processor, 12 Main memory storage area, 13
Shared memory area, 14 I / O dedicated processor (IO
P), 15 disks, 31 buffers, 32 buffer hashes, 33 buffer frame chains, 34 buffer pools, 41 buffer lock entries, 42 transactions.

Continuing on the front page (72) Inventor De Yang-hwa Republic of Korea Daejeon U-Sung Jung Min-dong Expo Apartment 406-204 (72) Inventor Jan Chul-Pak South Korea A) JP-A-4-2505041 (JP, A) JP-A-7-182213 (JP, A) JP-A 1-211140 (JP, A) JP-A-2-294835 (JP, A) JP-A-63 -311450 (JP, A) JP-A-5-134915 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) G06F 12/00 G06F 3/06

Claims (4)

(57) [Claims] At present, a transaction for executing a disk read operation includes a buffer lock entry used to lock a corresponding page, a buffer lock chain holding the buffer lock entry in a list form, and a buffer for a desired page. A buffer management structure including a buffer frame chain and a buffer hash anchor using an exclusive latch efficiently manages the simultaneous access of multiple users to the same disk page, thereby improving the performance of the storage system. in the buffer management method for, when a transaction accesses any page, after multiplication by exclusive latched in the buffer hash anchor by the transaction, the from the relevant buffer frame chain A first step of determining whether there is a buffer for the desired page; and determining the buffer lock chain only when the buffer for the desired page is not found from the corresponding buffer frame chain by the first step. A second step of determining whether a buffer lock entry for a desired page exists in the buffer lock chain; and, if a buffer for the desired page is found from the corresponding buffer frame chain based on the determination of the first step, the buffer After the exclusive latch of the buffer hash anchor is released and the buffer lock entry for the desired page is determined by the third step of using the page of the buffer and the second step. Is the buff If it exists in the buffer lock chain, the transaction information is registered in the channel of the corresponding buffer lock entry, the exclusive latch of the buffer hash anchor is released and the buffer is locked. A fourth step of applying a proper latch, and a fifth step of holding a lock on the desired page only when a buffer lock entry for the desired page does not exist in the buffer lock chain according to the determination of the second step. A buffer management method using a buffer lock technique in a multi-user environment storage system, characterized by: 2. The method of claim 5, further comprising: constructing a buffer lock entry for a corresponding page, connecting the buffer lock chain to the buffer lock chain, releasing an exclusive latch of the buffer hash anchor, and removing a free buffer. Reading the desired page from the disk after being allocated, latching the exclusive buffer of the buffer hash anchor and connecting the allocated buffer to the buffer frame chain; A step of finding a buffer lock entry from the buffer lock chain, separating the channel from the buffer lock entry, deleting the buffer lock entry from the buffer lock chain, and releasing a lock on the desired page. When the after releasing the exclusive latch buffer hash anchors, can be waked up (Wakeup) all transactions connected to said isolated channel,
2. The method according to claim 1, further comprising: performing the first process for the transaction in the execution state. 3. The method of claim 5, further comprising the step of not exclusively latching the buffer hash anchor while requesting allocation of the free buffer to read the desired page from the disk into the buffer. 3. The buffer management method according to claim 2, wherein a buffer lock technique is used in the multi-user environment storage system. 4. The multi-user environment storage system according to claim 1, wherein the buffer lock entry is used to hold a lock only for a page read from the disk without being found in a corresponding buffer frame chain. , A buffer management method using a buffer lock technique.